When in doubt, turn to the data (GDP and Median Household Income)

In the vein of the previous post, I will be analyzing data published by the government and available in the public domain to obtain a better understanding of issues important to this election. These posts have been motivated by the recent completion of Census 2010 and the impending election.

Today, we will compare GDP growth and the growth in median household income.

Information was obtained on GDP per capita (GDPC) at current market prices and median household income (MHI).

The charts below show that GDPC is positively correlated with MHI.

However, the growth in GDPC has far outpaced growth in MHI. For example, from 1999 to 2010, GDPC increased by 64.58%, while MHI increased by only 42.86%.

But let's be rigorous, lest I be accused of being selective about the base year in supporting my conclusions.

The table below shows the cumulative percentage increase in GDPC and MHI using each year between 1999 and 2010 as the base year. In each cell, the first percentage refers to GDPC and the second percentage (in parentheses) refers to MHI. Cells in which cumulative percentage increases in GDPC have outstripped MHI have been shaded red. Cells in which the opposite has occurred have been shaded blue.

There are a total of 46 red cells out of a total of 66 cells, meaning that MHI has on average fared worse than GDPC, base year effect or no.

Under the null hypothesis that each cell has 50-50 odds of being shaded blue or red, using a Gaussian approximation implies that such a result would occur less than 0.01% of the time if it were purely due to chance. Indeed, the test statistic is 3.2003, at the extreme right tail of the distribution.

But of course, such a statistical test is not truly valid, since data in each cell, being time-series data, is not independent of data in other cells. There is a dependency among the cells that violates the independence assumption of such a simple statistical test. Nonetheless, the results are illustrative.

Let's look at the data more closely. The chart below shows the annual rate of change (ARC) in GDPC and MHI for the years 2000 through to 2010.

It appears that the ARC for GDPC lies above MHI for 8 out of the 11 years. 

In addition, the AAGR for MHI lies above GDPC only in periods of recession (dot-com in 2000, the recent financial crisis). Why is this important? Because in periods of recession, the unemployment rate rises and unemployment disproportionately affects the lower income groups with below MHI. This causes less downside volatility in the MHI during a recession. In contrast, a strong economy creates full employment which tends to drive up incomes across the board.

In other words, MHI is more sensitive to economic booms than busts. This may account for why the AAGR for MHI is above the AAGR for GDPC only in times of recession.

I think I can safely say that MHI has lagged behind GDPC over the last 10 years.

But let's not stop there.

MHI refers to median household income. Household income may have increased over the years because workers in the household are being paid more. But it may also have increased because there are now more workers in each household, or workers are now working longer hours.

The government doesn't provide data on the median hourly wage. But census and household surveys helpfully provide data on the number of workers in each household (2000, 2005 and 2010), as well as the number of hours worked each week by each worker (2000, 2005, data not yet available for 2010). Time-series data on number of hours worked each week is available from MOM here, but I would prefer to stick to using census data for source consistency.

Data on number of workers in each household was grouped for large households, e.g. all households with 5 or more workers were grouped together. Data on number of persons in each household was similarly grouped for large households, e.g. all households with 8 or more persons were grouped together. I ignored the "more than" part when I calculated the weighted average number of workers and number of persons in each household.

The data is presented below. Note that year 2005 was problematic, as grouping was done for households with 4 or more workers, and 6 or more persons (as opposed to 5 or more workers and 8 or more persons, as was seen in years 2000 and 2010). This may have accounted for the break in trend in 2005 for weighted average number of workers per household. Nonetheless, there are more workers on average per household in year 2010 as compared to year 2000.

From the data, it appears that Singapore resident households are increasing in their average number of workers per household even as households are becoming smaller. This is unsurprising due to the increasing prevalence of two-income households and fewer kids.

For the data on number of hours worked each week, the number of hours was reported in bands. I took the mid-point of each band and calculated the weighted average number of hours worked per worker. For the extremely hard workers who work 65 or more hours per week, I assumed that the midpoint of this band was 72.5 hours. This is arbitrary, but unimportant as only a small percentage of the population works such long hours.

The weighted average number of hours worked per week was 47.47 in 2000 and 48.22 in 2005. Data is unavailable from Census 2010. But the two data points suggest an increase over time.

So, not only has MHI lagged GDPC, when adjusted for number of workers in each household and number of hours worked each week, the picture looks even worse.

But again, let's press on further.

One of the effects of the surge in housing prices, lacklustre growth in MHI and burgeoning income inequality is that people have increasingly moved to the outskirts of the island. Think Sengkang and Punggol where new flats have been built in the last 10 years. How have travel times to work changed?

Data on travel times to work is available for 2005 and 2010, but not for 2000.

Using the same methodology as above, I calculated a weighted average travelling time to work. For the extreme band of people who take more than 1 hour to get to work, I assumed a midpoint of 75 minutes. Note that in 2005, 5.09% of workers took more than 1 hour to get to work, while in 2010, 6.19% of workers fell into that group.

The weighted average travelling time was 30.33 minutes in 2005 and 31.81 minutes in 2010. Again, the Singapore worker is worse off.

I think I've made my case: The PAP has sucked at its job in improving the lot of Singaporeans as measured by MHI relative to GDP growth over the last 10 years.

I wish I had the data on annual remuneration for cabinet ministers over the last 10 years. That data I am confident will keep pace with GDP growth.